An introduction to nonparametric regression in demographic research

Abstract

"My intent in this...paper is to alert demographers to the existence of useful new methods, known collectively as nonparametric (NP) regressions, that have been made possible by reductions in computing costs over the last several decades. NP regressions use brute force computing power to free researchers from the need to make strong a priori assumptions about functional forms, thus allowing greater flexibility and accuracy in modelling relationships than in standard parametric approaches. Increased flexibility and computing power do not eliminate the central role of theory in guiding social science research, but they free empirical analysis from the requirement that relationships between variables must have mathematically convenient shapes....I begin this paper with a discussion of bias-variance tradeoffs in regression models, followed by a discussion of localized regression. Subsequent sections discuss specific NP techniques for bivariate and multivariate regressions, with examples. Throughout this paper I will illustrate with data from 17,851 mothers aged 15-45 who were interviewed in the 1988 U.S. National Maternal and Infant Health Survey." (SUMMARY IN FRE)